今回のファクター追加処理はさらだ側では一応完了だと思います。さらだ64側はこれからではありますが、新たなファクター込みのデータでの学習スタートさせてます。ここの所ではAutoMLはModel Builderには及ばない感じですがテストも兼ねて3時間学習中。今のAutoMLは時間プラス回数なんですが、3時間1回のみにしてます。その後、Model Builderでの学習に行きます。
このってか、AutoMLが終了予定17:25なので、そこからModel Builderになるので取敢えず12時間程度にして明日の朝には一旦様子が分かる様にして、その結果踏まえて少し考えたいと思います。まあ、その間にさらだ64のコーティングも進めておく予定です。
追記 2025.9.7 6:35
まず、AutoMLでの1回目はRMSE 1.4495で
| 579R | 1点 | 芝(315R) | ダート(244R) | 障害(20R) | 8頭以下(55R) | 9～12頭(163R) | 13頭以上(361R) | 多点 |
| 単勝 | 21.93% (68.57%) |
22.22% (64.95%) |
21.72% (71.76%) |
20.00% (86.50%) |
36.36% (74.91%) |
22.09% (64.29%) |
19.67% (69.53%) |
53.02% (73.95%) |
| 複勝 | 49.40% (74.47%) |
51.11% (74.51%) |
47.54% (74.59%) |
45.00% (72.50%) |
67.27% (85.09%) |
50.92% (69.63%) |
45.98% (75.04%) |
86.18% (81.45%) |
| 枠連 | 11.61% (75.87%) |
12.85% (81.16%) |
10.09% (70.22%) |
14.29% (73.57%) |
-- (--) |
14.72% (73.13%) |
9.14% (70.17%) |
24.64% (65.69%) |
| 馬連 | 10.19% (56.82%) |
11.75% (61.17%) |
7.79% (49.34%) |
15.00% (79.50%) |
32.73% (96.18%) |
11.04% (63.93%) |
6.37% (47.62%) |
21.24% (66.21%) |
| ワイド | 24.18% (102.78%) |
25.71% (77.78%) |
21.72% (135.53%) |
30.00% (97.00%) |
52.73% (98.73%) |
25.77% (73.01%) |
19.11% (116.84%) |
44.04% (85.61%) |
| 馬単 | 5.18% (58.41%) |
6.35% (70.83%) |
3.28% (37.91%) |
10.00% (113.00%) |
20.00% (98.55%) |
5.52% (68.10%) |
2.77% (47.92%) |
21.24% (61.57%) |
| 三連複 | 5.87% (84.75%) |
6.67% (46.89%) |
4.92% (126.60%) |
5.00% (170.50%) |
21.82% (78.18%) |
4.91% (65.89%) |
3.88% (94.27%) |
13.64% (49.41%) |
| 三連単 | 1.21% (44.75%) |
1.59% (32.98%) |
0.82% (63.61%) |
0.00% (0.00%) |
7.27% (94.00%) |
1.23% (62.70%) |
0.28% (29.14%) |
13.64% (44.24%) |
| 総合 | 51.30% (70.70%) |
53.65% (63.32%) |
48.77% (78.77%) |
45.00% (87.08%) |
74.55% (89.38%) |
52.76% (67.58%) |
47.09% (68.82%) |
86.70% (55.95%) |
って感じで、初回としてならまあまあ? で、Model Builder 12時間の1回目はRMSE 1.4652で
| 579R | 1点 | 芝(315R) | ダート(244R) | 障害(20R) | 8頭以下(55R) | 9～12頭(163R) | 13頭以上(361R) | 多点 |
| 単勝 | 19.17% (71.47%) |
15.87% (51.65%) |
24.59% (102.13%) |
5.00% (9.50%) |
30.91% (68.36%) |
22.70% (86.20%) |
15.79% (65.29%) |
46.29% (70.35%) |
| 複勝 | 46.63% (83.77%) |
43.81% (83.37%) |
50.82% (86.23%) |
40.00% (60.00%) |
50.91% (80.73%) |
52.15% (97.61%) |
43.49% (77.98%) |
81.87% (81.45%) |
| 枠連 | 11.61% (84.91%) |
12.45% (83.49%) |
9.65% (81.71%) |
28.57% (162.14%) |
-- (--) |
13.50% (59.45%) |
9.70% (88.64%) |
21.79% (75.93%) |
| 馬連 | 7.60% (81.43%) |
8.89% (60.92%) |
5.74% (103.57%) |
10.00% (134.50%) |
12.73% (34.36%) |
9.82% (56.87%) |
5.82% (99.70%) |
17.10% (74.42%) |
| ワイド | 19.00% (108.55%) |
20.63% (114.48%) |
16.39% (101.23%) |
25.00% (104.50%) |
40.00% (101.64%) |
22.70% (102.45%) |
14.13% (112.35%) |
35.92% (81.53%) |
| 馬単 | 3.11% (28.58%) |
3.17% (21.30%) |
2.87% (37.34%) |
5.00% (36.50%) |
10.91% (58.73%) |
4.91% (50.92%) |
1.11% (13.91%) |
17.10% (71.52%) |
| 三連複 | 4.15% (88.38%) |
4.76% (106.22%) |
3.69% (72.58%) |
0.00% (0.00%) |
14.55% (102.91%) |
5.52% (196.87%) |
1.94% (37.17%) |
10.54% (72.28%) |
| 三連単 | 1.04% (25.65%) |
1.27% (22.79%) |
0.82% (31.43%) |
0.00% (0.00%) |
3.64% (13.09%) |
1.84% (57.48%) |
0.28% (13.19%) |
10.54% (74.16%) |
| 総合 | 48.70% (71.33%) |
45.71% (67.61%) |
52.87% (76.99%) |
45.00% (59.55%) |
61.82% (65.69%) |
52.76% (88.48%) |
44.88% (63.53%) |
82.21% (74.45%) |
って感じなのは、やはりRMSE的にもAutoMLが少し良いからかな。なんにしてもどちらも初回だし、こっからどこまで精度があげられるかまた少し時間掛けて学習させてみます。ただ、さらだの方はあまり待たせてもと思うのでどこかで妥協して一旦リリースする予定です。
0 件のコメント:
コメントを投稿